Spectral Gradient: A Surface Reflectance Measurement Invariant to Geometry and Incident Illumination

Although photometric data is a readily available dense source of information in intensity images, it is not widely used in computer vision. A major drawback is its dependence on viewpoint and incident illumination. A novel methodology is presented which extracts reflectivity information of the various materials in the scene independent of incident light and scene geometry. A scene is captured under three different narrow-band color filters and the spectral derivatives of the scene are computed. The resulting spectral derivatives form a spectral gradient at each pixel. This spectral gradient is a surface reflectance descriptor which is invariant to scene geometry and incident illumination for smooth diffuse surfaces. The invariant properties of the spectral gradients make them a particularly appealing tool in many diverse areas of computer vision such as color constancy, tracking, scene classification, material classification, stereo correspondence, even re-illumination of a scene. Comments University of Pennsylvania Department of Computer and Information Science Technical Report No. MSCIS-99-02. This technical report is available at ScholarlyCommons: http://repository.upenn.edu/cis_reports/111 1 Spectral Gradient: A Surface Reflectance Measurement Invariant to Geometry and Incident Illumination Technical Report MS-CIS-99-02 Abstract Although photometric data is a readily available dense source of information in intensity images, it is not widely used in computer vision. A major drawback is its dependence on viewpoint and incident illumination. A novel methodology is presented which extracts reflectivity information of the various materials in the scene independent of incident light and scene geometry. A scene is captured under three different narrow-band color filters and the spectral derivatives of the scene are computed. The resulting spectral derivatives form a spectral gradient at each pixel. This spectral gradient is a surface reflectance descriptor which is invariant to scene geometry and incident illumination for smooth diffuse surfaces. The invariant properties of the spectral gradients make them a particularly appealing tool in many diverse areas of computer vision such as color constancy, tracking, scene classification, material classification, stereo correspondence, even re-illumination of a scene.Although photometric data is a readily available dense source of information in intensity images, it is not widely used in computer vision. A major drawback is its dependence on viewpoint and incident illumination. A novel methodology is presented which extracts reflectivity information of the various materials in the scene independent of incident light and scene geometry. A scene is captured under three different narrow-band color filters and the spectral derivatives of the scene are computed. The resulting spectral derivatives form a spectral gradient at each pixel. This spectral gradient is a surface reflectance descriptor which is invariant to scene geometry and incident illumination for smooth diffuse surfaces. The invariant properties of the spectral gradients make them a particularly appealing tool in many diverse areas of computer vision such as color constancy, tracking, scene classification, material classification, stereo correspondence, even re-illumination of a scene.